We compared two procedures for improving the social interactions of three autistic children. In a peer-initiation condition, confederates were taught to initiate interaction with the autistic children. In a teacher-antecedent condition, teachers prompted the autistic children to initiate with confederates, who had been taught to reciprocate. Using an alternating treatment design, differential effects were found. The peer-initiation procedure reliably increased the social responses of the autistic children, whereas the teacher-antecedent condition increased the initiations and responses of the autistic children. In addition, longer chains of social interaction occurred during the teacher-antecedent condition.